Input Files for TrainImagesRegression

Hi everyone!

I have a confusion with the input data in TrainImagesRegression (OTB-7.2.0).
I already read the cookbook, yet still don’t understand what to choose for ‘label image list’.
For an example case, I want to predict fractional vegetation cover (already in raster image) with the input of a multispectral image.
Really need explanation about the input files for this purpose.

Your attention and answer are really appreciated.

Thank you!


in the training application you need to provide two images:

  • a multispectral image containing the input values (Input predictor image)
  • a mono band image containing ground truth. The fractional vegetation cover in your case. (Input label image)

Several images can be provided, but the number of input predictor image must match the number of label image.

If I understand correctly your input image contains the predictor and predicted values in the same image. In this case you need to separate the input in two images, or you can use an extended filename to select the relevant bands:

Input predictor Image List : "image.tif?&bands=1:-2" #all bands except the last one
Input label Image List : "image.tif?&bands=-1" #last band

Hope that helps,

1 Like

Thanks a lot Cédric for your explanation!